Department of Electronics and Computer Engineering, Hanyang University, Seoul 04763, Republic of Korea.
ACS Appl Mater Interfaces. 2022 Oct 5;14(39):44724-44734. doi: 10.1021/acsami.2c08337. Epub 2022 Sep 27.
The conductive filament (CF) model, as an important means to realize synaptic functions, has received extensive attention and has been the subject of intense research. However, the random and uncontrollable growth of CFs seriously affects the performances of such devices. In this work, we prepared a neural synaptic device based on polyvinyl pyrrolidone-molybdenum disulfide quantum dot (MoS QD) nanocomposites. The doping with MoS QDs was found to control the growth mode of Ag CFs by providing active centers for the formation of Ag clusters, thus reducing the uncertainty surrounding the growth of Ag CFs. As a result, the device, with a low power consumption of 32.8 pJ/event, could be used to emulate a variety of synaptic functions, including long-term potentiation (LTP), long-term depression (LTD), paired-pulse facilitation, post-tetanic potentiation, short-term memory to long-term memory conversion, and "learning experience" behavior. After having undergone consecutive stimulation with different numbers of pulses, the device stably realized a "multi-level LTP to LTD conversion" function. Moreover, the synaptic characteristics of the device experienced no degradation due to mechanical stress. Finally, the simulation result based on the synaptic characteristics of our devices achieved a high recognition accuracy of 91.77% in learning and inference tests and showed clear digital classification results.
导电丝(CF)模型作为实现突触功能的重要手段,受到了广泛关注,并成为了研究的热点。然而,CF 的随机和不可控生长严重影响了这些器件的性能。在这项工作中,我们制备了一种基于聚乙烯吡咯烷酮-二硫化钼量子点(MoS QD)纳米复合材料的神经突触器件。研究发现,MoS QD 的掺杂通过为 Ag 团簇的形成提供活性中心,控制了 Ag CF 的生长模式,从而降低了 Ag CF 生长的不确定性。结果,该器件的功率消耗低至 32.8pJ/event,可用于模拟多种突触功能,包括长时程增强(LTP)、长时程抑制(LTD)、成对脉冲易化、强直后增强、短期记忆到长期记忆的转换,以及“学习经验”行为。在经过不同数量脉冲的连续刺激后,该器件稳定地实现了“多级 LTP 到 LTD 的转换”功能。此外,器件的突触特性不受机械应力的影响而退化。最后,基于我们器件的突触特性的模拟结果在学习和推理测试中实现了 91.77%的高识别准确率,并显示出清晰的数字分类结果。